Term 081

KFUPM
EE 656 – Robotics & Control
HW #5

2DOF Robotic Manipulator

Control Design & Simulation

By
Mohammad Shahab
227598

For
Dr. Ahmad Masoud

20 December 2008
 Term 081
EE 656 – Robotics & Control
HW #5

1) Robotic Arm Dynamics

We can put

𝑥1 = 𝐿1 sin 𝜃1

𝑦1 = 𝐿1 cos 𝜃1

𝑥2 = 𝐿1 sin 𝜃1 + 𝐿2 sin(𝜃1 + 𝜃2 )

𝑦2 = 𝐿1 cos 𝜃1 + 𝐿2 cos(𝜃1 + 𝜃2 )
So, Kinetic Energy could be formed as
1 1 1 1
𝐾𝐸 = 𝑀1 𝑥12 + 𝑀1 𝑦12 + 𝑀2 𝑥22 + 𝑀2 𝑦22
2 2 2 2
By simplification,

⟹
1 1 1
𝐾𝐸 = 𝑀1 + 𝑀2 𝐿21 𝜃12 + 𝑀2 𝐿22 𝜃12 + 𝑀2 𝐿22 𝜃1 𝜃2 + 𝑀2 𝐿22 𝜃22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 (𝜃1 𝜃2 + 𝜃12 )
2 2 2
And Potential Energy is

𝑃𝐸 = 𝑀1 𝑔𝐿1 cos 𝜃1 + 𝑀2 𝑔(𝐿1 cos 𝜃1 + 𝐿2 cos 𝜃1 + 𝜃2 )

So, by Lagrange Dynamics, we form the Lagrangian

ℒ = 𝐾𝐸 − 𝑃𝐸
1 1 1
ℒ= 𝑀1 + 𝑀2 𝐿21 𝜃12 + 𝑀2 𝐿22 𝜃12 + 𝑀2 𝐿22 𝜃1 𝜃2 + 𝑀2 𝐿22 𝜃22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 𝜃1 𝜃2 + 𝜃12
2 2 2
− 𝑀1 𝑔𝐿1 cos 𝜃1 − 𝑀2 𝑔 𝐿1 cos 𝜃1 + 𝐿2 cos 𝜃1 + 𝜃2
So, forming the dynamics equations to be

𝑑 𝜕ℒ 𝜕ℒ
𝑓𝜃1,2 = −
𝑑𝑡 𝜕𝜃1,2 𝜕𝜃1,2

So, the dynamic equations after simplifications become

𝑀1 + 𝑀2 𝐿21 + 𝑀2 𝐿22 + 2𝑀2 𝐿1 𝐿2 cos 𝜃2 𝜃1 + 𝑀2 𝐿22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 𝜃2

− 𝑀2 𝐿1 𝐿2 sin 𝜃2 2𝜃1 𝜃2 + 𝜃22 − 𝑀1 + 𝑀2 𝑔𝐿1 sin 𝜃1 − 𝑀2 𝑔𝐿2 sin 𝜃1 + 𝜃2
= 𝑓𝜃1

𝑀2 𝐿22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 𝜃1 + 𝑀2 𝐿22 𝜃2 − 𝑀2 𝐿1 𝐿2 sin 𝜃2 𝜃1 𝜃2 − 𝑀2 𝑔𝐿2 sin 𝜃1 + 𝜃2 = 𝑓𝜃2

So, we can describe the motion of the system by

𝐵 𝑞 𝑞 + 𝐶 𝑞 , 𝑞 + 𝑔(𝑞) = 𝐹
𝜃1
𝑞=
𝜃2

( 𝑀1 + 𝑀2 𝐿21 + 𝑀2 𝐿22 + 2𝑀2 𝐿1 𝐿2 cos 𝜃2 ) (𝑀2 𝐿22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 )

𝐵 𝑞 =
𝑀2 𝐿22 + 𝑀2 𝐿1 𝐿2 cos 𝜃2 𝑀2 𝐿22

−𝑀2 𝐿1 𝐿2 sin 𝜃2 2𝜃1 𝜃2 + 𝜃22

𝐶 𝑞, 𝑞 =
−𝑀2 𝐿1 𝐿2 sin 𝜃2 𝜃1 𝜃2
− 𝑀1 + 𝑀2 𝑔𝐿1 sin 𝜃1 − 𝑀2 𝑔𝐿2 sin 𝜃1 + 𝜃2
𝑔 𝑞 =
−𝑀2 𝑔𝐿2 sin 𝜃1 + 𝜃2

𝑓𝜃1
𝐹=
𝑓𝜃2

Here we have: 𝑀1 = 𝑀2 = 𝐿1 = 𝐿2 = 1
 2) Control Design
Having the system equation

𝐵 𝑞 𝑞 + 𝐶 𝑞, 𝑞 + 𝑔 𝑞 = 𝐹
We can have
−1
𝑞=𝐵 𝑞 [−𝐶 𝑞 , 𝑞 − 𝑔 𝑞 ] + 𝐹
With
−1
𝐹=𝐵 𝑞 𝐹⟺𝐹=𝐵 𝑞 𝐹
So, we decoupled the system to have the „new‟ (non-physical) input
𝑓1
𝐹=
𝑓2
However, the physical torque inputs to the system are

𝑓𝜃1 𝑓
=𝐵 𝑞 1
𝑓𝜃2 𝑓2

The error signals

𝑒 𝜃1 = 𝜃1𝑓 − 𝜃1

𝑒 𝜃2 = 𝜃2𝑓 − 𝜃2

With final positions

𝜋
𝜃1𝑓
= 2𝜋
𝜃2𝑓
−
2
The system has initial positions of
𝜋
−
𝜽𝟎 = 𝜋2
2
  PID Design
General structure of PID controller for any input would be

𝑓 = 𝐾𝑝 𝑒 + 𝐾𝐷 𝑒 + 𝐾𝐼 𝑒 𝑑𝑡

So, in our case,

𝑓1 = 𝐾𝑝1 𝜃1𝑓 − 𝜃1 − 𝐾𝐷1 𝜃1 + 𝐾𝐼1 𝑒 𝜃1 𝑑𝑡

𝑓2 = 𝐾𝑝2 𝜃2𝑓 − 𝜃2 − 𝐾𝐷2 𝜃2 + 𝐾𝐼2 𝑒 𝜃2 𝑑𝑡

So, the complete system equations with control would be

−1
𝑞=𝐵 𝑞 [−𝐶 𝑞 , 𝑞 − 𝑔 𝑞 ] + 𝐹
With

𝐾𝑝1 𝜃1𝑓 − 𝜃1 − 𝐾𝐷1 𝜃1 + 𝐾𝐼1 𝑒 𝜃1 𝑑𝑡

𝑓
𝐹= 1 =
𝑓2
𝐾𝑝2 𝜃2𝑓 − 𝜃2 − 𝐾𝐷2 𝜃2 + 𝐾𝐼2 𝑒 𝜃2 𝑑𝑡

Recall: with actual physical torques of

𝑓𝜃1 𝑓
=𝐵 𝑞 1
𝑓𝜃2 𝑓2

 Solution
In order to apply all controls of Proportional-Derivative-Integral actions, a „dummy‟ state is added for
each angle to resemble the integration inside the computer:

𝑥1 = 𝑒 𝜃1 𝑑𝑡 ⟹ 𝑥1 = 𝜃1𝑓 − 𝜃1

𝑥2 = 𝑒 𝜃2 𝑑𝑡 ⟹ 𝑥2 = 𝜃2𝑓 − 𝜃2
So, the complete system equations are
𝑥1 = 𝜃1𝑓 − 𝜃1
𝑥2 = 𝜃2𝑓 − 𝜃2
𝜃1 −1
𝐾𝑝1 𝜃1𝑓 − 𝜃1 − 𝐾𝐷1 𝜃1 + 𝐾𝐼1 𝑥1
=𝐵 𝑞 −𝐶 𝑞, 𝑞 − 𝑔 𝑞 +
𝜃2 𝐾𝑝2 𝜃2𝑓 − 𝜃2 − 𝐾𝐷2 𝜃2 + 𝐾𝐼2 𝑥2

In MATLAB, “ode45” command was used to solve the ODE. (Full program in Appendix)

By trial & error, the 2 controllers‟ parameters were tuned to have the best performance. The best values
for the parameters was found to be

𝐾𝑝1 = 15

𝐾𝐷1 = 7

𝐾𝐼1 = 10

𝐾𝑝2 = 15

𝐾𝐷2 = 10

𝐾𝐼2 = 10

Actually, by observing the structure of above ODE with control, we can see (roughly):

 𝐾𝑝 is related to direct error and to speed of evolution

 𝐾𝐷 is related to speed of interaction with change in states
 𝐾𝐼 is related to overall error cancelation

However, above arguments are rough because of the high nonlinearity of the equation that:

 produces sensitive interaction between controller components

 Small Changes in controller parameters would produce more overshoots and oscillations
 Controller parameters are highly sensitive to initial and final positions!
 So, online tuning of the PID should be considered for global system operation (i.e. non-fixed final
positions, trajectory tracking, etc.)
 3) States Results
Error forms of 𝜃1 & 𝜃2 is shown below

Figure 1: Errors waveforms of theta1 & theta2

Comments: You can see from above waveforms that:

 Acceptable overshoot
 Acceptable settling time
 „Linear‟ behavior

NOTE: The same design was tested on other initial and final positions, the result was very different.
 4) Torques Results
Here we analyze the torques resulted. The control inputs here are the torques. However, put in mind that
the actual joints torques are

𝑓𝜃1 𝑓
=𝐵 𝑞 1
𝑓𝜃2 𝑓2

The waveforms of joints torques are

Figure 2: Actual Torques on Joints

Comments: from above plots,

 𝜃1 -joint somehow encounter high torque in relatively small time
 Overall acceptable performance as relatively energy spent is O.K.

5) Motion Animation
Using MATLAB, the resulting motion is animated in xy-plane to show the actual motion of the robot.
An “mpg” file is generated. The file name is “2DOF_rob.mpg”. (MATLAB program is in Appendix)
 Appendix
1) System ODE function file (“r2dof.m”)
function xdot=r2dof(t,x,ths,spec,Kpid)

xdot=zeros(8,1);
%% set-points
th1s=ths(1);
th2s=ths(2);

%% Robot Specifications
M1=spec(3);
M2=spec(4);
L1=spec(1);
L2=spec(2);
g=9.8;

%% Inertia Matrix
b11=(M1+M2)*L1^2+M2*L2^2+2*M2*L1*L2*cos(x(4));
b12=M2*L2^2+M2*L1*L2*cos(x(4));
b21=M2*L2^2+M2*L1*L2*cos(x(4));
b22=M2*L2^2;
Bq=[b11 b12;b21 b22];

%% C Matrix
c1=-M2*L1*L2*sin(x(4))*(2*x(5)*x(6)+x(6)^2);
c2=-M2*L1*L2*sin(x(4))*x(5)*x(6);
Cq=[c1;c2];

%% Gravity Matrix
g1=-(M1+M2)*g*L1*sin(x(3))-M2*g*L2*sin(x(3)+x(4));
g2=-M2*g*L2*sin(x(3)+x(4));
Gq=[g1;g2];

%% PID Control
% PID parameters for theta 1
Kp1=Kpid(1);
Kd1=Kpid(2);
Ki1=Kpid(3);
% PID parameters for theta 2
Kp2=Kpid(4);
Kd2=Kpid(5);
Ki2=Kpid(6);
%decoupled control input
f1=Kp1*(th1s-x(3))-Kd1*x(5)+Ki1*(x(1));
f2=Kp2*(th2s-x(4))-Kd2*x(6)+Ki2*(x(2));
Fhat=[f1;f2];

F=Bq*Fhat; % actual input to the system

%% System states
xdot(1)=(th1s-x(3)); %dummy state of theta1 integration
xdot(2)=(th2s-x(4)); %dummy state of theta2 integration

xdot(3)=x(5); %theta1-dot
xdot(4)=x(6); %theta2-dot

q2dot=inv(Bq)*(-Cq-Gq+F);
xdot(5)=q2dot(1); %theta1-2dot
xdot(6)=q2dot(2); %theta1-2dot

%control input function output to outside computer program

xdot(7)=F(1);
xdot(8)=F(2);

2) System Solution and Simulation (“r2dof_cntrl.m”)

close all
clear all
clc
%% Initilization
th_int=[-pi/2 pi/2]; %initial positions
ths=[pi/2 -pi/2]; %set-points

x0=[0 0 th_int 0 0 0 0]; %states initial values

Ts=[0 20]; %time span

%% Robot Specifications
L1=1; %link 1
L2=1; %link 2
M1=1; %mass 1
M2=1; %mass 2
spec=[L1 L2 M1 M2];

%% PID Parameters
% PID parameters for theta 1
Kp1=15;
Kd1=7;
Ki1=10;
% PID parameters for theta 2
Kp2=15;
Kd2=10;
Ki2=10;

Kpid=[Kp1 Kd1 Ki1 Kp2 Kd2 Ki2];

%% ODE solving
% opt1=odeset('RelTol',1e-10,'AbsTol',1e-20,'NormControl','off');
[T,X] = ode45(@(t,x) r2dof(t,x,ths,spec,Kpid),Ts,x0);

%% Output
th1=X(:,3); %theta1 wavwform
th2=X(:,4); %theta2 wavwform

%torque inputs computation from the 7th,8th states inside ODE

F1=diff(X(:,7))./diff(T);
F2=diff(X(:,8))./diff(T);
tt=0:(T(end)/(length(F1)-1)):T(end);

%xy
x1=L1.*sin(th1); % X1
y1=L1.*cos(th1); % Y1
x2=L1.*sin(th1)+L2.*sin(th1+th2); % X2
y2=L1.*cos(th1)+L2.*cos(th1+th2); % Y2

%theta1 error plot

plot(T,ths(1)-th1)
grid
title('Theta-1 error')
ylabel('theta1 error (rad)')
xlabel('time (sec)')
%theta2 error plot
figure
plot(T,ths(2)-th2)
grid
title('Theta-2 error')
ylabel('theta2 error (rad)')
xlabel('time (sec)')
%torque1 plot
figure
plot(tt,F1)
grid
title('Torque of theta 1')
ylabel('theta1 torque')
xlabel('time (sec)')
%torque2 plot
figure
plot(tt,F2)
grid
title('Torque of theta 2')
ylabel('theta2 torque')
xlabel('time (sec)')

3) Robot Animation (“movieHW5.m”)

%% setting frames speed
d=2;
j=1:d:length(T);

%% generating images in 2D
figure
for i=1:length(j)-1
hold off
plot([x1(j(i)) x2(j(i))],[y1(j(i)) y2(j(i))],'o',[0 x1(j(i))],[0
y1(j(i))],'k',[x1(j(i)) x2(j(i))],[y1(j(i)) y2(j(i))],'k')
title('Motion of 2DOF Robotic Arm')
xlabel('x')
ylabel('y')
axis([-3 3 -3 3]);
grid
hold on
MM(i)=getframe(gcf);
end
drawnow;

%% exporting to 'mpg' movie

mpgwrite(MM,'RGB','2DOF_rob.mpg')